This is amazing!

(I may have gotten a little too into mapping this out in my head...)

I noticed the arbitrariness of just multiplying the three values together too. I had a similar thought to Mike: multiplying assumes the factors are independent in some sense, right? Which I’m not sure would strictly hold here.

I wonder if you could turn your three factors into features and fit a simple statistical model based on human-labeled data? That way, you could let the model calculate the weights for AC, SS, and EF — instead of relying on an arbitrary multiplication. What do you think? Would that work?